From legal sources to knowledge: building the EuroLegalBot dataset

Abstract

A legal AI assistant is only as reliable as the knowledge on which it is built. With Deliverable D2.2 – Dataset, EuroLegalBot moves from methodology to implementation, creating the structured knowledge base that supports the chatbot’s understanding of the European Arrest Warrant (EAW).

The dataset brings together legal material from official European and national sources and organises it in formats suitable for the training and evaluation of the system. The initial knowledge hub contains 6,339 files, classified according to institutional origin, jurisdiction and territorial scope.

The work also pays particular attention to transparency and quality. The initial training process includes a set of 92 judicial cases, while the wider dataset is cleaned, validated and structured to improve the relevance and reliability of the information retrieved by EuroLegalBot.

D2.2 represents an essential step in transforming authoritative legal information into a usable knowledge resource for AI-supported judicial cooperation.